public static Tensor cat(Tensor[] x, int dim)
{
var dtype = x[0].dtype;
var data = new object[x.Length];
var shapes = new int[x.Length][];
var strides = new int[x.Length][];
var y_shape = new int[x[0].__shape.Length];
for (int i = 0; i < x[0].__shape.Length; i++)
{
y_shape[i] = x[0].__shape[i];
}
var requires_grad = false;
for (int i = 0; i < x.Length; i++)
{
var t = x[i];
if (t.requires_grad)
{
requires_grad = true;
}
shapes[i] = t.__shape;
strides[i] = t.__strides;
if (i > 0)
{
if (t.__shape.Length != y_shape.Length)
{
throw new TorchException("TorchException: Invalid tensor sizes for torch.cat: the number of dimensions must be the same.");
}
for (int j = 0; j < t.__shape.Length; j++)
{
if (j == dim)
{
y_shape[j] += t.__shape[j];
continue;
}
if (t.__shape[j] != y_shape[j])
{
throw new TorchException(string.Format("TorchException: Invalid tensor sizes for torch.cat: impossible to match shapes {0} and {1} of dimention {2} of tensors {3} and {4}.", t.__shape[j], y_shape[j], j, i - 1, i));
}
}
}
switch (t.dtype)
{
case torch.uint8:
{
if (dtype == torch.@bool)
{
throw new TorchException("TorchException: unable to cat bool tensor and non-bool tensor.");
}
data[i] = t.__uint8;
break;
}
case torch.int8:
{
if (dtype == torch.@bool)
{
throw new TorchException("TorchException: unable to cat bool tensor and non-bool tensor.");
}
data[i] = t.__int8;
if (dtype == torch.uint8)
{
dtype = torch.int8;
}
break;
}
case torch.int16:
{
if (dtype == torch.@bool)
{
throw new TorchException("TorchException: unable to cat bool tensor and non-bool tensor.");
}
data[i] = t.__int16;
if (@sizeof(dtype) == 1)
{
dtype = torch.int16;
}
break;
}
case torch.int32:
{
if (dtype == torch.@bool)
{
throw new TorchException("TorchException: unable to cat bool tensor and non-bool tensor.");
}
data[i] = t.__int32;
if (@sizeof(dtype) < 4)
{
dtype = torch.int32;
}
break;
}
case torch.int64:
{
if (dtype == torch.@bool)
{
throw new TorchException("TorchException: unable to cat bool tensor and non-bool tensor.");
}
data[i] = t.__int64;
if (@sizeof(dtype) == 8)
{
dtype = torch.int64;
}
break;
}
case torch.half:
{
if (dtype == torch.@bool)
{
throw new TorchException("TorchException: unable to cat bool tensor and non-bool tensor.");
}
data[i] = t.__half;
if (!dtype.is_floating_point())
{
switch (@sizeof(dtype))
{
case 1:
case 2:
{
dtype = torch.float16;
break;
}
case 4:
{
dtype = torch.float32;
break;
}
case 8:
{
dtype = torch.float64;
break;
}
}
}
break;
}
case torch.float32:
{
if (dtype == torch.@bool)
{
throw new TorchException("TorchException: unable to cat bool tensor and non-bool tensor.");
}
data[i] = t.__float;
if (!dtype.is_floating_point())
{
switch (@sizeof(dtype))
{
case 1:
case 2:
case 4:
{
dtype = torch.float32;
break;
}
case 8:
{
dtype = torch.float64;
break;
}
}
}
else
{
if (dtype == torch.float16)
{
dtype = torch.float32;
break;
}
if (dtype == torch.float64)
{
dtype = torch.float64;
break;
}
}
break;
}
case torch.float64:
{
if (dtype == torch.@bool)
{
throw new TorchException("TorchException: unable to cat bool tensor and non-bool tensor.");
}
data[i] = t.__double;
dtype = torch.float64;
break;
}
case torch.@bool:
{
if (dtype != torch.@bool)
{
throw new TorchException("TorchException: unable to cat bool tensor and non-bool tensor.");
}
data[i] = t.__bool;
break;
}
}
}
var y = new Tensor(y_shape, dtype, (!torch.autograd.grad_mode.no_grad.prev) && requires_grad);
switch (dtype)
{
case torch.float16:
{
MKL.Cat(data, shapes, strides, dim, y.__half, y.__shape, y.__strides);
break;
}
case torch.float32:
{
MKL.Cat(data, shapes, strides, dim, y.__float, y.__shape, y.__strides);
break;
}
case torch.float64:
{
MKL.Cat(data, shapes, strides, dim, y.__double, y.__shape, y.__strides);
break;
}
case torch.int8:
{
MKL.Cat(data, shapes, strides, dim, y.__int8, y.__shape, y.__strides);
break;
}
case torch.uint8:
{
MKL.Cat(data, shapes, strides, dim, y.__uint8, y.__shape, y.__strides);
break;
}
case torch.int16:
{
MKL.Cat(data, shapes, strides, dim, y.__int16, y.__shape, y.__strides);
break;
}
case torch.int32:
{
MKL.Cat(data, shapes, strides, dim, y.__int32, y.__shape, y.__strides);
break;
}
case torch.int64:
{
MKL.Cat(data, shapes, strides, dim, y.__int64, y.__shape, y.__strides);
break;
}
case torch.@bool:
{
MKL.Cat(data, shapes, strides, dim, y.__bool, y.__shape, y.__strides);
break;
}
}
return(y);
}
